I'm new to transformer models and have a basic question that I didn't find an answer to in the docs... with many traditional modeling techniques, we attempt to make the target stationary (e.g. by subtracting the mean). Should we do this when using TimeGPT as well?... or would doing so be a bad idea given that the model is trained on non-stationary time-series data to start with. Similarly, we usually transform and normalize independent variables prior to modeling (in econometrics and ML). Should we do the same for exogenous features when using TimeGPT? Thank you.

Hello! Those transformations are not needed with TimeGPT. The model was trained on minimally preprocessed data such that you don't have to do those steps when using TimeGPT.

Cool, thanks Marco.